CN112617999A

CN112617999A - Guiding cardiac ablation using Machine Learning (ML)

Info

Publication number: CN112617999A
Application number: CN202011001118.8A
Authority: CN
Inventors: M·阿米特; L·措雷夫; Y·A·阿莫斯; A·沙吉
Original assignee: Biosense Webster Israel Ltd
Current assignee: Biosense Webster Israel Ltd
Priority date: 2019-09-22
Filing date: 2020-09-22
Publication date: 2021-04-09
Also published as: KR20210036278A; RU2020130901A3; RU2020130901A; EP3795078A1; IL277420A; JP2021049341A; US20210085387A1

Abstract

The invention is entitled "guiding cardiac ablation using Machine Learning (ML)". A system includes an interface and a processor. The interface is configured to receive data characterizing an initial ablation operation applied to a region of a heart of a patient. If found necessary, the processor is configured to automatically specify a supplemental ablation operation to be applied to the region based on the received data.

Description

Guiding cardiac ablation using Machine Learning (ML)

Cross Reference to Related Applications

This patent application claims the benefit of us provisional patent application 62/903,850 filed 2019, 9, 22, the disclosure of which is incorporated herein by reference.

Technical Field

The present invention relates generally to the processing of electrophysiological signals and ablation, and in particular, to optimizing cardiac ablation parameters using Machine Learning (ML).

Background

Methods and systems for planning and guiding ablation procedures based on patient data to treat arrhythmias have previously been reported in the patent literature. For example, U.S. patent No. 9,463,072 describes a method and system for patient-specific planning and guidance of electrophysiological interventions. A patient-specific anatomical heart model is generated from cardiac image data of a patient. A patient-specific cardiac electrophysiology model is generated based on a patient-specific anatomical heart model of the patient and the patient-specific electrophysiology measurements. Virtual electrophysiological interventions are performed using a patient-specific cardiac electrophysiology model. A simulated Electrocardiogram (ECG) signal is calculated in response to each virtual electrophysiological intervention. Embodiments of the present invention utilize advanced machine learning algorithms, LBM-EP (lattic-Boltzmann method for electrophysiology) techniques for near real-time modeling of cardiac electrophysiology, and a model of the generation of ECG signals to predict and display patient-specific electrocardiograms after virtual EP therapy.

As another example, U.S. patent No. 9,277,970 describes a method and system for patient-specific planning and guidance of ablation protocols for cardiac arrhythmias. A patient-specific anatomical heart model is generated based on the pre-operative cardiac image data. The patient-specific anatomical heart model is registered to a coordinate system of an intra-operative image acquired during an ablation procedure. One or more ablation site guide maps are generated based on the registered patient-specific anatomical heart model and intra-operative patient-specific measurements acquired during the ablation procedure. The ablation site guidance map may include a myocardial spread and action potential duration map. An ablation site guide map is generated using a computational model of cardiac electrophysiology that is personalized by fitting parameters of the cardiac electrophysiology model using intra-operative patient-specific measurement results. Displaying, by a display device, the ablation site guidance map during the ablation procedure. In one embodiment, registering the patient-specific anatomical heart model to the intraoperative three-dimensional rotational angiography image acquired during the ablation procedure includes using a machine learning algorithm to calculate a probabilistic map of the pericardium in the three-dimensional rotational angiography image.

Disclosure of Invention

Embodiments of the invention described below provide a system that includes an interface and a processor. The interface is configured to receive data characterizing an initial ablation operation applied to a region of a heart of a patient. If found necessary, the processor is configured to automatically specify a supplemental ablation operation to be applied to the region based on the received data.

In some embodiments, the processor is configured to designate the supplemental ablation by: assessing a quality of the initial ablation operation, and designating the supplemental ablation operation in response to finding that the quality of the initial ablation operation does not satisfy a quality criterion.

In some embodiments, the data characterizing the initial ablation operation includes at least one of: lesion (depth) depth; the radius of the focal zone; a long shaft of the ablation stove; a short axis of the ablation focus; 3D position of the ablation focus; dissecting a focus; and focal surface area.

In one embodiment, in designating the supplemental ablation operation, the processor is configured to designate a location of repeated ablation.

In another embodiment, in designating the supplemental ablation operation, the processor is configured to indicate a gap in a segment of ablation points.

In yet another embodiment, in designating the supplemental ablation operation, the processor is configured to designate in real-time that additional ablations are to be performed in the vicinity of segments of ablation points.

In some embodiments, in specifying the supplemental ablation operation, the processor is further configured to specify values of one or more ablation parameters to be used in the supplemental ablation.

In some embodiments, the data characterizing the initial ablation operation includes at least one of: body surface Electrocardiogram (ECG) signals; changes in body surface ECG signals; intracardiac ECG signals; changes in intracardiac ECG signals; an impedance of the ablation electrode; a change in impedance of the ablation electrode; the temperature of the ablated tissue; a change in temperature of the ablated tissue; force on the ablated tissue; a change in force on the ablated tissue; an ablation catheter type; 3D position of ablation points; predicted anatomical locations of ablation points; ablation duration of the ablation point; the rate of flushing; and power delivered during ablation.

In other embodiments, the data characterizing the initial ablation operation includes one or both of: changes in ultrasound reflectance of ablated tissue, and changes in Magnetic Resonance Images (MRI) of ablated tissue.

In some embodiments, the processor is configured to automatically specify the supplemental ablation operation by applying a trained Machine Learning (ML) model.

In one implementation, the ML model includes at least one of an auto-encoder, a variational auto-encoder, a generic countermeasure network (GAN), a Random Forest (RF), supervised ML, and enhanced ML.

There is additionally provided, in accordance with another embodiment of the present invention, a method including receiving data characterizing an initial ablation operation applied to a region of a heart of a patient. If found necessary, the processor automatically specifies a supplemental ablation operation to be applied to the region based on the received data.

Drawings

The invention will be more fully understood from the following detailed description of embodiments of the invention taken together with the accompanying drawings, in which:

fig. 1 is a schematic illustration of a catheter-based Electrophysiological (EP) sensing, signal analysis, and IRE ablation system according to an exemplary embodiment of the present invention;

fig. 2 illustrates a depth learning algorithm for lesion estimation based on an automatic encoder and Random Forest (RF) according to an exemplary embodiment of the present invention;

FIG. 3 is a flow diagram of the training and inference use of a Machine Learning (ML) model for estimating and correcting ablation results, according to an alternative exemplary embodiment of the present invention; and is

FIG. 4 is a screen shot of a visualization of Pulmonary Vein Isolation (PVI) planned using the ML model of FIG. 3, according to an exemplary embodiment of the present invention.

Detailed Description

SUMMARY

Cardiac ablation is a common procedure for treating cardiac arrhythmias by forming lesions in the cardiac tissue of a patient. Such lesions may be formed by irreversible electroporation (IRE) or using other types of ablation energy such as Radio Frequency (RF), both of which may be applied using a catheter. In IRE ablation, the catheter is manipulated so that an electrode disposed on the distal end of the catheter is in contact with the tissue. Then, a high voltage bipolar pulse is applied between the electrodes, and a strong electric field pulse generated in the tissue causes cell death and lesion generation. In RF ablation, an alternating RF current is applied to the tissue through one or more electrodes, causing cell death by heat.

Generally, in ablation procedures in a heart chamber for correcting arrhythmias, it is important to achieve contiguity and transmurality (sufficient lesion depth) in the ablation. The absence of either typically results in the wavefront "leaking" through the ablated tissue. To check for leakage, for example in a procedure for achieving Pulmonary Vein Isolation (PVI), a physician paces the heart (i.e., injects a signal into the heart) on one side of the ablation line in order to stimulate the heart, and checks whether the signal is present on the other side. If the signal is not present, electrical isolation has been achieved as expected. However, if the signal does appear, the physician typically adds an ablation point.

Parameters, such as ablation line contiguity index (ACLI), may be defined for scoring the contiguity and transmurality of the ablation line. However, such parameters may be estimated only after the ablation procedure has been completed, so that the PVI may only be checked after a complete cycle has been performed.

Thus, a typical current workflow may have the following steps:

first "ablation cycle" of PV isolation "

2. Testing via stimulation

3. Ablating tissue to close the gap based on the stimulus; and/or

2. Testing via arrhythmia-inducing drugs (e.g., adenosine)

3. Ablating tissue to close the gap

It will be appreciated that the adenosine challenge step 2 described above may not be implemented in a daily clinical workflow, as it is typically used in research settings.

Even if isolation is achieved during the procedure, the arrhythmia may return at some later date. In this case, a "rerun" of the procedure may be necessary.

Embodiments of the present invention described below provide a system and Machine Learning (ML) method for predicting the success of a cardiac ablation procedure based solely on information acquired during treatment (e.g., acquired using the ablation catheter itself), as described below. The prediction is achieved by (i) estimating lesion characteristics, such as the extent to which a lesion penetrates the wall, in some embodiments, and (ii) estimating the level of contiguity and the level of transmurality in a lesion in other embodiments. Later embodiments may guide the physician in real time in the event additional ablation points are needed.

In some embodiments, a processor receives data characterizing an initial ablation operation applied to a region of a heart of a patient. If found necessary, the processor automatically specifies a supplemental ablation operation to be applied to the region based on the received data. For example, the processor evaluates the quality of the initial ablation operation and designates a supplemental ablation operation in response to finding that the quality of the initial ablation operation does not meet a quality criterion (e.g., realized contiguity and/or transmurality).

Embodiments of the present invention may be used to provide recommendations for a focused source and a Repetitive Active Pattern (RAP). Embodiments of the present invention can also be used to provide an estimate of ablation quality as well as a "best" ablation strategy for persistent AF drivers and maintenance, RAP, focused and fibrotic tissue.

In one embodiment of the invention, a Machine Learning (ML) model, such as an Artificial Neural Network (ANN), is generated. The ANN model is trained using initial ablation data including body surface Electrocardiogram (ECG) signals, intracardiac ECG (icecg) signals (also known as Electrograms (EGMs)), 3D location information of the collected data, and ablation parameters including power for ablation, time period of ablation, temperature measured during ablation, and impedance of the catheter electrodes performing the ablation. Other parameters used to train the model include, but are not limited to, the catheter used, the forces measured by the catheter, and changes in parameters such as temperature and impedance. Once the model has been generated, the model uses the values of at least some of these parameters to arrive at the results formulated by the model.

In some embodiments, a base event derived from clinical data and preclinical data of ablation training data is used for training. Such data may include actual lesion parameters obtained at a range of ablation powers, including parameters such as surface area and depth of tissue necrosis.

When performing ablation in a new patient, the processor uses the ML model to estimate the ablated foci (e.g., its radius and depth), which is provided to the physician. These values are typically provided on a Graphical User Interface (GUI) that may also provide visualization of the ablation.

In other embodiments of the invention, the trained ML model identifies locations after the first ablation cycle that are potential candidates for ablation "rerun".

Further, just before and/or during a new ablation, the processor may use the model in order to predict the level of contiguity and transmurality in the ablation using the ablation data for a particular patient (i.e., any data acquired during the ablation procedure).

Using the above ML model enables the physician to reduce the time spent on the procedure by predicting the outcome of the ablation procedure compared to the time used for the current workflow described above. The model enables the physician to create effective isolation by ablating lines (i.e., lines with high contiguity and transmural scores in the first round of the ablation procedure). Thus, using the model allows for a simpler, faster, and more efficient process compared to prior art systems.

Although an ANN model is used as an example herein, one skilled in the art can select from other ML models available, such as decision tree learning, Support Vector Machines (SVMs), and bayesian networks. ANN models include, for example, convolutional NN (CNN), autoencoder, and Probabilistic Neural Network (PNN). Typically, the processor or processors used (hereinafter collectively referred to as "processors") are programmed in software containing specific algorithms that enable the processor to perform each of the processor-related steps and functions listed above. Typically, training is performed using a computing system that includes multiple processors, such as a Graphics Processing Unit (GPU) or Tensor Processing Unit (TPU). However, any of these processors may also be a Central Processing Unit (CPU).

The ability to assess lesion parameters (e.g., diameter, depth) and contiguity and transmurality in ablation in real time based on the limited data described above for ML algorithm inference allows for a simple assessment of the quality of the ablation treatment, can result in a more accurate ablation profile, and generally results in an improvement in the outcome of the ablation procedure.

Description of the System

Fig. 1 is a schematic illustration of a catheter-based Electrophysiological (EP) sensing, signal analysis, and IRE ablation system 20 according to an embodiment of the present invention. System 20 can be, for example, manufactured by Biosense-Webster (Irvine, California)

And 3, system. As shown, system 20 includes a catheter 21 having a shaft 22 that is navigated by a physician 30 into a heart 26 (inset 25) of a patient 28. In the illustrated example, the physician 30 inserts the shaft 22 through the sheath 23 while affixing with the proximal end of the catheterA proximal manipulator 32 manipulates the shaft 22.

In the embodiments described herein, the catheter 21 may be used for any suitable diagnostic purpose and/or tissue ablation, such as electrophysiological mapping and IRE ablation of the heart 26, respectively. ECG recorder 35 may receive various types of ECG signals sensed by system 20 during the procedure.

As shown in the inset 25, the distal end of the shaft 22 of the catheter 21 is provided with a multi-electrode basket catheter 40. Inset 45 shows the arrangement of the plurality of electrodes 48 of basket catheter 40. The proximal end of catheter 21 is connected to console 24 for transmission of an electrogram, for example, taken by electrodes 48.

Console 24 includes a processor 41 (typically a general purpose computer) having suitable front end and interface circuitry 38 for receiving EP signals (e.g., ECG signals) from electrodes 48 of catheter 21 as well as non-EP signals, such as position signals. To this end, processor 41 is connected to electrode 48 via a wire extending within shaft 22. The interface circuit 38 is further configured to receive ECG signals, such as ECG signals from a multi-channel (e.g., 12-lead) ECG device, which may be an ECG recorder 35, and non-ECG signals from surface body electrodes 49. Typically, the electrodes 49 are attached to the skin around the chest and legs of the patient 28. Processor 41 is connected to electrode 49 by wires extending through cable 39 to receive signals from electrode 49.

Four of the body surface electrodes 49 are named according to standard ECG protocols: MA (right arm), LA (left arm), ML (right leg), and LL (left leg). The Wilson Central Terminal (WCT) may be formed by three of the four named body surface electrodes 49 and the resulting ECG signal V_WCTReceived by the interface circuit 38.

During the EP mapping procedure, the position of the electrodes 48 is tracked while the electrodes are within the patient's heart 26. For this purpose, electrical signals are transmitted between the electrodes 48 and the body surface electrodes 49. Based on the signals, and given the known positions of the electrodes 22 on the patient's body, the processor 41 calculates an estimated position of each electrode 22 electrode within the patient's heart. This tracking may be performed using an active current position (ACL) system manufactured by Biosense-Webster (Irvine, California) as described in U.S. patent No. 8,456,182, the disclosure of which is incorporated herein by reference.

Thus, the processor may associate any given signal (such as an EGM) received from the electrode 48 with the location at which the signal was acquired. Processor 41 uses the information contained in these signals to construct an EP map, such as a Local Activation Time (LAT) map, for presentation on a display. In the illustrated embodiment, using an algorithm including an ML algorithm applied to EP and other data (e.g., irrigation rate), processor 41 estimates the contiguity and transmurality of the lesions ablated by the system as shown in fig. 2 and 3.

To perform IRE ablation, electrodes 48 are connected (e.g., switched) to an IRE pulse generator 47 in console 24 that includes processor-controlled switching circuitry (e.g., a relay array, not shown). Using the estimates provided by the disclosed techniques (such as the level of ablation contiguity), processor 41 or the physician may select the electrodes (via the switching circuitry) to be connected to pulse generator 37 to apply the IRE pulses.

During IRE ablation, the initial ablation data defined below can be used for inferences made by one of the ML models described above to further evaluate (e.g., in real-time) the lesion parameters, as depicted in fig. 2, as well as the contiguity and transmurality of a set of lesions.

The processor 41 is typically programmed in software to perform the functions described herein. For example, the software may be downloaded to the processors in electronic form, over a network, or alternatively or in addition to, the software may be provided and/or stored on non-transitory tangible media, such as magnetic, optical, or electronic memory. In particular, processor 41 executes a dedicated algorithm, such as that included in fig. 3, as disclosed herein, which enables processor 41 to perform the steps disclosed herein, as further described below.

Using ML to guide cardiac ablation

Fig. 2 is a diagram of the operation of a Machine Learning (ML) model according to an exemplary embodiment of the present invention. The model is constructed from a random forest regression tree based on the ablation training data described below. The random forest classifier 204 is a committee of decision trees in which each decision tree has been fed a subset of the attributes of the data and predictions are made based on that subset. The mode of the actual predicted values of the decision tree are considered to provide the final random forest answers 208 and 210. Random forest classifiers generally mitigate overfitting that exists in independent decision trees, resulting in more robust and accurate classifiers.

The model is based on a deep learning auto-encoder and a random forest regression tree. The auto-encoder performs dimensionality reduction on a set of features u of the feature space that is later used as lesion assessment.

The auto-encoder 202 includes two parts: an encoder and a decoder. The encoder maps the input (here the ECG signal and/or the IcECG signal) to the hidden representation (u) via a non-linear transformation. The decoder then maps the hidden representation back to the reconstructed data via another non-linear transformation:

u＝f(IcECG,θ_{encoder for encoding a video signal}),

Embodiments of the present invention use the same network architecture for ECG and IcECG reconstruction. The L2 normalization function is minimized to learn a set of θ_{Encoder for encoding a video signal}、

Weights to reconstruct the IcECG (or body surface ECG). An auto-encoder is implemented using a fully connected convolutional neural network (FCN) with a predetermined number of layers of encoders and decoders. Random forest regression is then performed based on the encoded representation u, patient history (e.g., AF duration, NYHA score) parameters, patient demographics (e.g., age, BMI), and ablation characteristics (e.g., power, temperature profile) in order to predict the ablation lesion depth.

In this embodiment, the model uses the ablation feature space as an input layer, i.e., as input to a random forest. Ablation feature space refers to ablation characteristics (e.g., power, impedance drop, stability, ablation index, and x, y, z position of each ablation point of cardiac tissue). Each ablation point includes those features as a time series of sixty (60) samples per second, so the time-varying nature of each ablation point is also modeled and used as part of the ablation feature space.

The anatomy of each ablation point is also part of the input space of the model. For example, if a point is predicted to be part of a right Wide Area Circumferential Ablation (WACA), the ML model for WACA classifies the ablation site as one of the following:

1. lower right

2. Right back

3. Upper right part

4. Right front

For the left WACA, each ablation site is associated with one of:

1. left lower part

2. Left back

3. Left front

4. Upper left of

5. Ridge

An ML model for classifying Ablation sites is provided in U.S. provisional patent application No. 63/059060 entitled "Automatic Segmentation of Anatomical Structures with Wide Area Circumferential Ablation Points," filed on 30/7/2020 (assigned to the assignee of the present patent application).

The disclosed model provides as output an estimate of the lesion surface area (as a radius) 210 and the lesion depth 208, and is operated on by the computer processor 206. In operation, the processor applies an algorithm to the constructed model, including inputting data from an ablation procedure, and outputting lesion estimates 208 and 210.

In this embodiment, the underlying scene of the ablation training data used is derived from clinical data and preclinical data. Some data were calculated from in vivo chest opening protocols performed on pigs or/and sheep. In a protocol, lesions are created at a range of power in order to achieve different lesion depths in both the atrium and ventricle in sheep and pigs. The surface area and depth of tissue necrosis were collected.

Furthermore, it is possible to deliver energy at a range of power to achieve different focal depths in a human subject. ultrasound/MRI (magnetic resonance imaging) can be used to measure the energy delivered to both the atrium and ventricle as well as the surface area and depth of necrosis.

For both animal and human subjects, ablation training data collection includes ECG and intracardiac ECG signals, ablation catheter type (e.g., focus, lasso, basket, balloon), 3D location of ablation points, ablation duration for each point, whether irrigation (and if so, irrigation rate) is used, impedance of the ablation electrodes, power delivered, and temperature profile measured during ablation. Additional optional data includes intracardiac ultrasound, external ultrasound, real-time CT, and real-time MRI images.

The data mentioned above is used as input to a random forest regression tree model.

As shown in fig. 2, the model estimates the depth and area of tissue necrosis by two output nodes including the average depth of the foci and the surface area of the foci (as the radius of the foci). In an alternative embodiment, a model is generated with three output nodes, where the major and minor axes of the ellipse are used to estimate the surface area of the lesion.

In all embodiments, the estimated surface area of necrosis after each ablation is displayed in real time.

Fig. 3 is a flow diagram of the training and inference use of a Machine Learning (ML) model for estimating and correcting ablation results according to an alternative embodiment of the present invention. In particular, the embodiment described in fig. 3 helps to decide whether and where ablation re-progress is needed in the heart chamber.

In one embodiment of the algorithm, the model may be used to identify potential locations for ablation re-progress (i.e., repeated ablations) after a first "ablation cycle".

In the disclosed embodiments of the present invention, the model uses a fully connected neural network with an ablated feature space as an input layer. The network has two hidden layers, which contain the corrective linear units (relus), and a single output neuron-the first classifier that supports binary-including ablation success or recurrence. A gradient descent optimizer is used to estimate the weights of the network to minimize cross-entropy loss.

In the event that the first classifier predicts a re-progression, the second classifier marks a "low depth" small surface "lesion as a potential location for re-progression of ablation.

In another embodiment of the algorithm (the algorithm is activated after several ablations but before the first round of PVI is completed), the algorithm identifies potential gaps in the segment of ablation points, e.g., when there are 50mm segments near to each other ablation points and one of the ablations has a low impedance drop and the catheter is 20mm from the segment, the system informs the physician of these potential gaps.

In yet another embodiment of the algorithm, the system typically notifies in "real-time" during ablation whether there is a potential gap within the previous ablation or whether additional ablations should be performed near the current ablation point.

In another embodiment of the algorithm, the system indicates to the physician where the next ablation point should be performed. The system also lists the ablation parameters to be used and their values until the ablation is completed. The algorithm typically determines whether completion has been achieved and notifies the physician accordingly.

According to one presented embodiment, the algorithm is divided into two parts: algorithm preparation 101 and algorithm usage 102.

The algorithm prepares to perform a process starting with the ML modeling step 70 to generate an ML algorithm for estimating the ablation result. Such models may be supervised or enhanced ML models, variational autocodes and generic countermeasure networks (GANs), among other possible options. The model accepts as input the EP and ablation results, as well as other inputs described below.

Next, at an ML algorithm training step 72, the processor trains the algorithm using a database that includes ablation training data and base live results.

In some embodiments, the ablation training data for the model is divided into two categories: data from a first ablation session that achieves immediate success, and data from a first ablation session that requires a re-procedure (e.g., after finding that the quality of the initial ablation operation does not meet a quality criterion, such as contiguity achieved).

In disclosed embodiments, ablation training data may include information obtained from actual treatment with a system (such as CARTO), for example:

1. ablation catheter type

2. 3D position of ablation points

3. Anatomical location of ablation points

4. Power for ablation

5. Duration of spot ablation

6. Rinsing

7. Catheter stability, i.e. the force applied to the catheter during ablation

8. A parameter related to the area of ablation for validating transmural ablation based on the "predicted" tissue width.

9.12 lead ECG, or any type of body surface ECG and IcECG

10. Tissue response, e.g. temperature, ultrasound reflection change, ECG signal reduction, impedance change

11. External device-only data, e.g. MRI and/or ultrasound data

12. External device data combined with any one of items 1 to 10.

In alternative disclosed embodiments, the ablation training data may include a slave-based profile

Training data for the generated images (or similar systems). Such images include:

1. the generated image sets based on different CARTO maps LAT, voltages, visitags (e.g., spherical markers on the image, as shown in fig. 4. sphere size and color indicate power delivered and time delivered), etc., for each of the protocol phases (left PV and right PV) and for different phases of ablation.

2. Images taken in several fixed views.

3. The start of training, includes left PV isolated and right PV isolated images.

The second stage of training, includes segment (part of the complete PVI) isolated images.

4. All map images with the same view, and images used for training including all CARTO staining methods: voltage, LAT, and bipolar maps

5. Images of all Ablation tag options typically generated for machine learning only (including standard CARTO Visitag/SurPoints/exposure Index and additional maps presenting other parameters, such as by coloring the Visitag ball based on catheter stability during Ablation or by coloring the Visitag ball based on any other parameter that may contribute to machine learning):

standard CARTO ablation Point

Points with a color gradient according to any of the following parameters:

o energy supplied

Force of o catheter

O a catheter angle

A previous catheter position

O ablation power

O ablation time

O' a temperature

ICEG bipolar pressure drop

O catheter stability

Respiration during ablation

Erzu wash

In any of the disclosed embodiments, the correction of the first PVI cycle may also be used as part of the training data. These corrections may include the following:

a. parameters of additional ablation points, such as ablation catheter type, 3D location of ablation points, power used for ablation, duration of point ablation, irrigation, catheter stability, catheter force, and/or:

b. as an image of the ablation performed as a correction. If the image is taken immediately after the re-run, the image may include the ablation location and the supplied energy.

In addition to the training data described above, the base live data is also used to construct the model, as shown. The underlying live data is typically based on a hospital database and can be divided into two parts:

a. immediate success, as determined by pacing and/or applying adenosine challenges, so that short-term follow-up is not required.

A 12 month follow-up success, determined by reviewing rerun cases and cases with follow-up information and based on clinical studies including follow-up.

In some embodiments, the basic live data may include validity success and clinical success criteria, as in Mansour M et al entitled "ablation of persistent atrial fibrillation by contact force sensing catheter: expected multicenter rule testing (Persistent identification association with contact force sensing transmitter: The proactive multi center PRECEPT Trial) "and in JACC: clinical electrophysiology, defined in the study of rules published in 8 months, 2020, vol.6, 8.

In some embodiments, the base live data may include acquired data that is not presented to a physician because it does not have a known clinical benefit.

While the above description refers to embodiments that include two separate machine learning models, it should be understood that the embodiments may be combined into one model that is implemented to perform the functions of both embodiments.

At a trained model storage step 74, the algorithm preparation ends by storing the trained model in a non-transitory computer readable medium, such as a keyboard (memory stick). In an alternative embodiment, the model is sent in advance and its optimization parameters (such as weights for the ANN) are sent separately after training.

The algorithm uses 102 to perform a process that begins at an algorithm upload step 76 during which the user uploads the entire ML model or its optimization parameters (e.g., weights) to the processor. Next, a processor, such as processor 28, receives patient data (e.g., ECG and EGM) similar to the type of data used in the training from

electrodes

49 and 48, respectively, at a patient data receiving step 78.

Next, using the trained ML model for inference, the processor inputs data from the selected patient to the model and implements an algorithm on the model such that the model can output supplemental ablation operations (e.g., corrective actions) needed to perform more contiguous ablation, for example, at an ablation recommendation step 80. After installation to the processor, the trained model may be used with multiple patients.

The exemplary flow chart shown in fig. 3 was chosen solely for conceptual clarity. This embodiment may also include additional steps of the algorithm, such as receiving an indication of the degree of physical contact of the electrodes with the tissue being diagnosed. This and other possible steps have been purposely omitted from the disclosure herein in order to provide a more simplified flow diagram.

FIG. 4 is a screen shot of a visualization of Pulmonary Vein Isolation (PVI) planned using the ML model of FIG. 3, according to an embodiment of the present invention.

FIG. 4 depicts the output of an "evaluation engine" (e.g., the ML algorithm of FIG. 3) that estimates the contiguity and transmurality of PVIs. In one embodiment, during the training phase, the assessment engine has a set of conditions that are labeled by a trained physician with the following information:

1. ablation anatomical site (left WACA, right WACA, top line, etc.)

2. Type of ablation (active, inactive ablation, immediate reconnection or reconnection of ablation points)

3. Width and depth of ablation

Based on the feature space described above, the ML algorithm provides a prediction of the effectiveness of each ablation (e.g., the probability between zero and one), the width and depth of the lesion. The engine may also recommend regions of potential immediate reconnection sites or long term (re-) reconnection sites.

The disk 402 shown in the figure represents an ablation point, and dimensions (and/or disk gray scale) are automatically created by the evaluation engine to depict the contiguity or transmurality of PVIs. The size of the disc may represent ablation effectiveness probabilities, they may also represent the width and/or depth of the lesion.

In some embodiments, the engine may also give suggestions of ablation regions 404 to avoid a rerun condition.

Although the embodiments described herein relate primarily to cardiac ablation applications, the methods and systems described herein may also be used for other medical applications, such as renal denervation, after retraining with relevant data inputs and taking into account relevant success criteria.

It will thus be appreciated that the embodiments described above are cited by way of example, and that the present invention is not limited to what has been particularly shown and described hereinabove. Rather, the scope of the present invention includes both combinations and subcombinations of the various features described hereinabove, as well as variations and modifications thereof which would occur to persons skilled in the art upon reading the foregoing description and which are not disclosed in the prior art. Documents incorporated by reference into this patent application are considered an integral part of the application, except that definitions in this specification should only be considered if any term defined in these incorporated documents conflicts with a definition explicitly or implicitly set forth in this specification.

Claims

1. A system for guiding cardiac ablation, the system comprising:

an interface configured to receive data characterizing an initial ablation operation applied to a region of a heart of a patient; and

a processor configured to automatically specify a supplemental ablation operation to be applied to the region based on the received data.

2. The system of claim 1, wherein the processor is configured to designate the supplemental ablation by: assessing a quality of the initial ablation operation, and designating the supplemental ablation operation in response to finding that the quality of the initial ablation operation does not satisfy a quality criterion.

3. The system of claim 1, wherein the data characterizing the initial ablation operation comprises at least one of:

the depth of the ablation focus;

the radius of the focal zone;

a long shaft of the ablation stove;

a short axis of the ablation focus;

3D position of the ablation focus;

dissecting a focus; and

the surface area of the ablation focus.

4. The system of claim 1, wherein in specifying the supplemental ablation operation, the processor is configured to specify a location of repeated ablations.

5. The system of claim 1, wherein in specifying the supplemental ablation operation, the processor is configured to indicate a gap in a segment of ablation points.

6. The system of claim 1, wherein in specifying the supplemental ablation operation, the processor is configured to specify in real-time that additional ablations are to be performed in proximity to segments of ablation points.

7. The system of claim 1, wherein in specifying the supplemental ablation operation, the processor is further configured to specify values of one or more ablation parameters to be used in the supplemental ablation.

8. The system of claim 1, wherein the data characterizing the initial ablation operation comprises at least one of:

body surface Electrocardiogram (ECG) signals;

changes in body surface ECG signals;

intracardiac ECG signals;

changes in intracardiac ECG signals;

an impedance of the ablation electrode;

a change in impedance of the ablation electrode;

the temperature of the ablated tissue;

a change in temperature of the ablated tissue;

force on the ablated tissue;

a change in force on the ablated tissue;

an ablation catheter type;

3D position of ablation points;

predicted anatomical locations of ablation points;

ablation duration of the ablation point;

the rate of flushing; and

power delivered during ablation.

9. The system of claim 8, wherein the data characterizing the initial ablation operation comprises one or both of:

a change in ultrasonic reflectance of ablated tissue; and

a change in a Magnetic Resonance Image (MRI) of ablated tissue.

10. The system of claim 1, wherein the processor is configured to automatically specify the supplemental ablation operation by applying a trained Machine Learning (ML) model.

11. The system of claim 10, wherein the ML model comprises at least one of an auto-encoder, a variational auto-encoder, a generic countermeasure network (GAN), a Random Forest (RF), supervised ML, and enhanced ML.

12. A method for guiding cardiac ablation, the method comprising:

receiving data characterizing an initial ablation operation applied to a region of a heart of a patient; and

automatically specifying, by the processor, a supplemental ablation operation to be applied to the region based on the received data.

13. The method of claim 12, wherein designating the supplemental ablation comprises: assessing a quality of the initial ablation operation, and designating the supplemental ablation operation in response to finding that the quality of the initial ablation operation does not satisfy a quality criterion.

14. The method of claim 12, wherein the data characterizing the initial ablation operation comprises at least one of:

the depth of the ablation focus;

the radius of the focal zone;

a long shaft of the ablation stove;

a short axis of the ablation focus;

3D position of the ablation focus;

dissecting a focus; and

the surface area of the ablation focus.

15. The method of claim 12, wherein designating the supplemental ablation operation comprises designating a location of a repeated ablation.

16. The method of claim 12, wherein designating the supplemental ablation operation comprises indicating a gap in a segment of ablation points.

17. The method of claim 12, wherein designating the supplemental ablation operation comprises designating, in real-time, additional ablations to be performed in the vicinity of segments of ablation points.

18. The method of claim 12, wherein specifying the supplemental ablation operation comprises specifying values for one or more ablation parameters to be used in the supplemental ablation operation.

19. The method of claim 12, wherein the data characterizing an initial ablation operation comprises at least one of:

body surface Electrocardiogram (ECG) signals;

changes in body surface ECG signals;

intracardiac ECG signals;

changes in intracardiac ECG signals;

an impedance of the ablation electrode;

a change in impedance of the ablation electrode;

the temperature of the ablated tissue;

a change in temperature of the ablated tissue;

force on the ablated tissue;

a change in force on the ablated tissue;

an ablation catheter type;

3D position of ablation points;

predicted anatomical locations of ablation points;

ablation duration for each point;

the rate of flushing; and

power delivered during ablation.

20. The method of claim 19, wherein the data characterizing the initial ablation operation comprises one or both of:

a change in ultrasonic reflectance of ablated tissue; and

a change in a Magnetic Resonance Image (MRI) of ablated tissue.

21. The method of claim 12, wherein automatically specifying the supplemental ablation operation comprises applying a trained Machine Learning (ML) model.

22. The method of claim 21, wherein the ML model comprises at least one of an auto-encoder, a variational auto-encoder, a generic countermeasure network (GAN), a Random Forest (RF), supervised ML, and enhanced ML.